Algorithms for quadratic constrained matrix problems (Q1200862)

The authors discuss the quadratic constrained matrix problem, i.e. the problem of computing the best possible estimate \(X\) of an unknown matrix with known bounds on individual entries, known row and column totals, and known weighted totals of subsets of individual entries. To solve this problem the authors propose a method that breaks the above problem on three classes of subproblems that are easy to solve: a row subproblem, a column subproblem, and a cut-set subproblem. Each of these in turn is an equilibrium problem that is efficiently solvable by specially designed equilibration operators. This general approach may be thought of as a block-cyclic ascent method applied to the dual problem, or as a modified Gauss-Seidel scheme to solve the Kuhn-Tucker conditions. Outer loop and the alternative subproblem algorithms are presented as well as results of computational experiments.